Author: David, Trend Research

Trend Overview: Cerebras (CBRS) released its first quarterly report post-IPO; Q1 core revenue was $191 million, a 92% year-on-year increase, exceeding market expectations. However, Q2 core gross margin guidance plummeted from 46.5% to 36%-38%, causing stock price to drop over 10% post-market. This company, which makes chips from a whole wafer and bets on the AI inference track, holds a contract with OpenAI worth over $20 billion and a partnership framework with AWS, with annual revenue guidance of $855 million to $865 million. The growth data looks solid, but the valuation controversy is significant.

Core Focus Points

1. Revenue exceeded expectations, guidance also surpassed expectations. Q1 core revenue was $191.3 million (YoY +92%), above the consensus expectation of about $181 million. Annual core revenue guidance is $855 million to $865 million (YoY +69%), exceeding the market expectation of $828 million. Under GAAP standards, cloud and service revenue was $82.8 million, a 178% year-on-year increase, the fastest-growing segment.
2. The sharp drop in gross margin guidance is the biggest negative this season. Q1 core gross margin was 47%, an increase of nearly 5 percentage points year-on-year. However, Q2 guidance dropped to 36%-38%, a decline of about 10 percentage points from Q1; annual guidance is 38%-41%. Management attributed the cause to insufficient data center capacity: the company is temporarily leasing back systems from existing customers of sold hardware to deploy capacity, which worsens short-term costs. The stock price fell over 10% post-market.
3. Customer concentration shows improvement direction, but is far from resolved. In fiscal year 2025, 86% of revenue comes from two UAE-related entities (MBZUAI accounts for 62%, G42 for 24%). OpenAI began contributing revenue in February 2026, and the partnership with AWS is expected to reflect in finances by 2027. True revenue diversification will not be validated until 2027.
4. Valuation pricing corresponds to 2028. At a post-market price of about $200, CBRS corresponds to about 90 times the revenue of the past twelve months; even using the median of the annual guidance of $865 million, the forward P/S remains above 50 times. Among the 10 covering analysts, the median target price is $300 (range of $250-$340), implicitly assuming that the OpenAI contract over $20 billion and AWS deployment are fulfilled on time and in quantity.
5. Short-term catalysts and constraints coexist. Catalysts: Acceleration of OpenAI's 750MW computing power deployment, AWS inference solutions implemented, new data center capacity coming online in the second half of the year. Constraints: Lock-up period includes unconventional early release clauses (triggered if market value exceeds $40 billion, current market value is near that threshold), unclear path for gross margin recovery, OpenAI itself has not yet turned a profit and has already scaled back some computing power commitments.

The financial report exposes the transformation of the business model: from selling chips to selling computing power

The easiest aspect to be overlooked in the Q1 financial report is the change in revenue structure.

On a core basis, hardware revenue was $111.6 million, accounting for 58% of total revenue; cloud and service revenue was $79.8 million, accounting for 42%. A year ago, this ratio was roughly 70:30. Cloud service revenue increased 167% year-on-year, nearly three times the growth rate of hardware.

Management made this trend clearer in the conference call:

In the upcoming quarters, hardware revenue will phase out, as the company will allocate more hardware capacity to its cloud to fulfill the inference power contracts with OpenAI and AWS, rather than selling directly to customers. Cerebras is transitioning from "a chip-selling company" to "a computing power-selling company."

This transformation also directly explains why the gross margin sharply declined in Q2.

In the conference call, an analyst probed the details of capacity deployment, and management revealed:

The current bottleneck of the company is not in chip supply from TSMC, but in the physical space of the data center. To expedite the delivery of computing power to OpenAI, Cerebras is "temporarily renting back" already sold hardware systems from G42 (its previous largest customer and also a minority stake investor).

Leasing third-party facilities to deploy its systems will worsen cost structure in the short term; this is the main reason gross margin guidance dropped from 47% to 36%-38%. Management provided a timeline stating that new data centers would start coming online in the second half of the year; at this point, cost pressures will ease.

The financial structure of the OpenAI contract is also worth dissecting. On the surface, it is a multi-year computing power procurement contract worth over $20 billion, but beneath this lie three layers of relationships intertwined: OpenAI provided Cerebras with a $1 billion operating capital loan (reflected in the Q1 balance sheet as $621 million in current loans and $362 million in non-current loans), while also obtaining warrants for Cerebras' stock.

In other words, OpenAI simultaneously acts as Cerebras' largest customer, creditor, and potential shareholder. The risk warning in S-1 states that if Cerebras fails to deliver capacity as promised, OpenAI has the right to terminate the contract and trigger loan repayment.

The cooperation framework with AWS adopts a "split inference" architecture: AWS's Trainium-3 chip is responsible for processing prompt input (prefill stage), while Cerebras' CS-3 system is specialized for high-speed output generation (decode stage). This design allows Cerebras not to bear the complete inference chain, focusing only on the segment where it has the greatest speed advantage. However, management refused to disclose the specific scale of the AWS collaboration during the Q&A session and stated that revenue contributions would only be reflected in financials by 2027.

Both large contracts share a common feature: huge contract volumes, but long realization paths, and highly dependent on the progress of Cerebras' data center construction.

The annual revenue guidance of $855 million to $865 million implies that the last three quarters need to average about $220 million, with growth accelerating each quarter. Management stated that "the year-on-year growth rate for each quarter in 2026 will increase, with more revenue concentrated in the second half."

Bullish Logic: Nine Investment Banks Simultaneously Recommend Buying, What Are They Betting On?

On June 8, the first day after the end of the IPO silent period, nine underwriters simultaneously initiated coverage, all giving buy or add ratings. CBRS rose by 18.3% in a single day. This collective bullish stance is not uncommon among new stocks in the U.S. market (underwriters have inherent interest connections), but their betting logic points to the same core proposition.

Proposition One: The battlefield for AI computing power is shifting from training to inference, and the competition rules for inference scenarios differ from training.

Joseph Moore, a Morgan Stanley analyst, gave an add rating and a target price of $250 in his first coverage report on June 8. His core argument is that training scenes compete on total computing power throughput, in which NVIDIA's GPU clusters dominate; inference scenes compete on responsiveness speed and latency, as models have to handle millions of user requests per second, and speed directly affects service costs and user experience. Cerebras' wafer-level chip has a structural advantage in inference latency because its on-chip SRAM capacity far exceeds that of conventional GPUs, eliminating frequent data movement to external storage. Moore stated that Cerebras is "the only company commercializing wafer-level processors," creating a first-mover advantage over NVIDIA.

Citi analyst Atif Malik provided the highest target price of $340 among the coverage. Mizuho added a technical detail in their June 8 report: the WSE-3 chip has 44GB of SRAM built-in, several times that of Google's latest TPU and Groq LPU; this hardware-level difference cannot be bridged through architectural optimization in the short term.

Proposition Two: Two large contracts push Cerebras from a "technology story" to an "revenue story."

The OpenAI contract exceeds $20 billion, covering 750MW of inference power with multi-year delivery. Based on a five-year amortization calculation, this contract alone will contribute about $4 billion in revenue per year, nearly five times the median of the 2026 annual revenue guidance. Although management declined to disclose the specific amount for the AWS partnership, the framework has been confirmed: Cerebras' inference capability will be available to global enterprise clients through Amazon Bedrock.

Q1 financial data provides early validation. OpenAI began deploying Cerebras systems in February, and cloud service revenue jumped from less than $30 million in the same quarter last year to nearly $80 million. Management stated that "the year-on-year growth rate for each quarter in 2026 will increase, with more revenue concentrated in the second half," with annual guidance of $855 million to $865 million exceeding the consensus expectation of $828 million.

Proposition Three: The density of coverage after the silent period is itself a signal.

The median target price among 10 analysts is $300, with a minimum of $250 (Morgan Stanley) and a maximum of $340 (Citi). Based on a post-market price of $200, the median target price implies about 50% upside potential. Wedbush (target price $270), Needham ($300), Barclays ($280), TD Cowen ($275), Craig-Hallum (buy) all initiated coverage within the same week.

The underlying assumption of the bullish logic can be summarized in one sentence:

If AI inference becomes a larger computing power market than training (several institutions predict that inference computing expenditures will exceed those of training by 2027), and if Cerebras' speed advantage is real and sustainable, then it only needs to capture 3%-5% of a market where NVIDIA occupies over 80% share to sustain its current valuation.

Bearish Logic: Gross Margin, Customer Concentration, and the Fragility of a $50 Billion Valuation

The three propositions of the bulls have corresponding rebuttals from the bears.

Rebuttal One: The moat of speed advantage in inference may be narrower than imagined.

Cerebras' speed advantage is based on on-chip SRAM capacity, but NVIDIA is not standing still. The B300 chip released by NVIDIA in March significantly increased HBM bandwidth, while Groq's LPU architecture is also rapidly iterating in inference scenarios.

From another perspective: Cerebras' customers are currently highly concentrated in OpenAI and AWS, while OpenAI is also one of NVIDIA's largest GPU purchasing customers, and AWS's self-developed Trainium chip is covering more and more inference scenarios. The major customers of Cerebras are simultaneously betting on alternatives, which suggests that its speed premium will continue to face pricing pressure.

Rebuttal Two: The decline in gross margin may not only be "temporary."

Management attributed the drop in Q2 gross margin from 47% to 36%-38% to temporary leasing costs due to insufficient data center capacity. However, this explanation rests on the premise that "costs will improve after new data centers come online in the second half of the year."

Considering that revenue scale will rise sharply in the second half (management clearly stated revenue will load towards the backend), and the ramp-up of new data center capacity also requires time and capital investment, this recovery path will not be easy.

A deeper question concerns the impact of the business model transition on gross margin. As Cerebras transitions from selling hardware to selling cloud computing power, it means taking on the costs of construction, operation, and depreciation of data centers. With depreciation from self-built data centers being accounted for, whether cloud service gross margins can be maintained above 50% remains uncertain. The profit margin ceiling of this business model has yet to be tested.

Rebuttal Three: Customer concentration is a "renamed but unresolved" issue.

In 2024, G42 alone contributed 85% of Cerebras' revenue. In 2025, G42's share dropped to 24%, but MBZUAI (Mohammed bin Zayed University of Artificial Intelligence) soared from nothing to 62%. The S-1 prospectus clearly marks these two as "related parties." Together, the two UAE-related entities still account for 86% of revenue. The diversification of revenue sources is more about a change of names rather than substantial dispersion.

Finally, the lock-up period for CBRS's IPO contains an unconventional clause:

If the company's market value remains above $40 billion, insiders' shares can be released early. At the post-market price of $200, the current market value is about $45 billion, already close to the trigger line. Regarding short positions, as of May 29, the short ratio was 17.15% of float shares, a relatively high level. Once the lock-up period is released early, leading to a large number of insider shares becoming available for sale, combined with existing short-selling pressure, the stock price may face concentrated selling.